Granted Patent B2

US 12,373,415 · App. 18/207,050 · Granted Jul 29, 2025

Inventors: Yeye He (Bellevue, WA); Cong Yan (Issaquah, WA); Yue Wang (Redmond, WA); Surajit Chaudhuri (Kirkland, WA); Peng Li (Atlanta, GA)

Assignee: Microsoft Technology Licensing, LLC

G06F16/2282G06F16/21

Abstract

Claims (43)

1. A method comprising:

obtaining a conforming relational data table that conforms to a relational format;

generating a training example by performing a first transformational operation on the conforming relational data table, the first transformational operation resulting in a non-conforming data table that is formatted differently than the relational format;

identifying a second transformational operation that will transform the non-conforming data table back to the conforming relational data table;

training a machine learning model with the training example to obtain a trained machine learning model, the training involving evaluating a loss function to determine weights of the trained machine learning model, the loss function being based at least on whether the machine learning model correctly predicts the second transformational operation that will transform the non-conforming data table back to the conforming relational data table;

synthesizing, with the trained machine learning model, a program for a different non-conforming data table that is formatted differently than the relational format; and,

transforming, with the synthesized program, the different non-conforming data table into another conforming relational data table that conforms to the relational format.

2. The method of claim 1 , further comprising:

selecting the first transformational operation from a set of transformational operators.

3. The method of claim 1 , wherein selecting the first transformational operation comprises selecting a single transformational operator or multiple serially performed transformational operators.

4. The method of claim 3 , wherein identifying the second transformational operation comprises identifying a single inverse transformational operator or multiple serially performed inverse transformational operators.

5. The method of claim 1 , wherein the obtaining comprises obtaining multiple conforming relational data tables and training examples are generated from each of the multiple conforming relational data tables.

6. The method of claim 5 , wherein the training the machine learning model comprises training the machine learning model utilizing the training examples generated from the multiple conforming relational data tables.

7. The method of claim 1 , wherein training the machine learning model comprises training the machine learning model without a human-provided label for the training example.

8. The method of claim 1 , wherein the machine learning model is trained prior to receiving the different non-conforming data table.

9. The method of claim 8 , further comprising:

generating a user interface; and

receiving input identifying the different non-conforming data table through the user interface.

10. The method of claim 9 , further comprising:

presenting the transforming of the different non-conforming data table into the another conforming relational data table on the user interface.

11. The method of claim 1 , the trained machine learning model comprising an embedding layer and a convolutional layer.

12. A system, comprising:

a processor; and,

a storage resource storing computer-readable instructions which, when executed by the processor, cause the processor to:

obtain conforming relational data tables that are formatted in a relational format;

generate training data without human labelling by identifying:

a transformational operator that will transform an individual conforming relational data table to a non-conforming data table that is formatted differently than the relational format, and

an inverse transformational operator that will transform the non-conforming data table back to the individual conforming relational data table that is formatted in the relational format; and,

train a machine learning model with the training data by evaluating a loss function to determine weights of the trained machine learning model, the loss function being based at least on whether the machine learning model correctly predicts the inverse transformational operation that will transform the individual non-conforming data table back to the conforming relational data table.

13. The system of claim 12 , wherein the processor is further configured to synthesize programs with the trained machine learning model for other individual conforming relational data tables.

14. The system of claim 13 , wherein the processor is further configured to rank the synthesized programs.

15. The system of claim 14 , wherein the processor is further configured to re-rank the synthesized programs with input-output re-ranking.

16. The system of claim 15 , wherein the processor is further configured to receive an additional data table and utilize the trained machine learning model to synthesize a program to transform the additional data table into a conforming relational data table.

17. The system of claim 16 , wherein the processor is further configured to cause a user interface to be generated and to receive the additional data table via the user interface.

18. A computing device, comprising:

hardware; and,

an Auto-Tables component configured to utilize a trained machine learning model to synthesize a program to transform a particular input data table into a particular conforming relational data table,

wherein, prior to synthesizing the program, the trained machine learning model has been trained by evaluating a loss function to determine weights of the trained machine learning model, the loss function being based at least on whether the machine learning model correctly predicts operations that will transform other non-conforming data tables into other conforming relational data tables,

wherein the particular conforming relational data table is in a relational format, and

wherein the particular input data table is in a different format than the relational format.

19. The computing device of claim 18 , wherein the Auto-Tables component is further configured to cause a user interface to be generated and to receive input identifying the particular input data table via the user interface.

20. The computing device of claim 19 , wherein the Auto-Tables component is further configured to cause the transformation of the particular input data table into the particular conforming relational data table to be presented on the user interface.

21. The computing device of claim 18 , wherein the Auto-Tables component is further configured to recognize that another input data table is already a conforming relational data table and to not transform the another input data table.

Assignments (1)

Continuity (1)

References Cited (82)